CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: fix bug in topk-moe softmax (#16711)

Browse Source
This commit is contained in:
Aman Gupta
2025-10-22 12:33:08 +08:00
committed by GitHub
parent 03792ad936
commit 9285325ce0
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
ggml/src/ggml-cuda/topk-moe.cu
View File

@@ -141,7 +141,7 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *
wt_sum = warp_reduce_sum(wt_sum); wt_sum = warp_reduce_sum(wt_sum);
const float inv_sum = 1.0f / wt_sum; const float inv_sum = 1.0f / wt_sum;
for (int i = threadIdx.x; i < n_expert_used; i += WARP_SIZE) { for (int i = 0; i < experts_per_thread; i++) {
output_weights[i] *= inv_sum; output_weights[i] *= inv_sum;
} }
} }